Daily Trivia - How Smart is GPT-4 vs Real Human in understanding common knowledge?-

Daily Trivia - How Smart is GPT-4 vs Real Human in understanding common knowledge?-

Very seldom we see a tech news grow so fast outside of its domain in just hours. OpenAI ChatGPT manage to turn the whole humankind attention to the possibility of AGI (Artificial General Intelligent) in the next few year.

You have see a lot of news about how the latest GPT-4 model is way better than GPT-3.5 and start having good result in exam.

However, have you even wonder, how smart in GPT-4 vs real life human in understanding common knowledge and reasoning?

Meet HellaSwag. Basically a test created by researcher in University of Washington to test whether you can use common knowledge to reason and finish the sentence.

So far, REAL HUMAN have been scoring around 94-96.5 mark in HellaSwag. And There wasn’t any AI model that can score higher than 86 mark all these while, not even OpenAI GPT-3.5, Meta LLaMa. Human still easily win in all those tricky question on common knowledge.

GPT-4 come. And drop a 95.3 mark. Which mean GPT-4 is already smarter in common knowledge reasoning than quite a number of REAL HUMAN.

People might get excited about GPT-4 taking over SAT/BAR exam, but people should look at this result, and start thinking, maybe…just maybe I can replace human generalist to do some of the general knowledge stuff?

Last but not least, remember GPT-4 is kinda already done 8 months ago and they spend 8 month just to test and audit its safety/risk assessment. Who know, GPT-5 might already in the itinerary at this moment.

I hope you learn something new today :D